Paste an episode. Get it back in Spanish, in your own voice.
Studio is our own speech, language, and voice models — Melo, Percept, and Noema — as creator tools. Generate voice, transcribe and subtitle audio, and dub a script into another language, all from one credit balance.
Say this in another language. One flow, one file.
Translate the script, render it in a Melo voice, and stitch the pieces into a single downloadable WAV. Not three tools you operate by hand — one flow in the console.
“Welcome back to the show. Today we’re looking at how small teams ship faster.”
“Bienvenidos de nuevo al programa. Hoy veremos cómo los equipos pequeños avanzan más rápido.”
Split on sentences, never mid-sentence
The script is chunked on sentence boundaries — Latin, CJK, and Arabic/Urdu punctuation all handled — so the voice never drops cadence at a seam.
Only languages we can actually speak
The dub languages come from the voice catalog itself, so every option in the list has a voice behind it. No dead ends.
Three models. All of them ours.
Melo — voice
- Text to a downloadable WAV, with speed control
- 5,000 characters per render — longer scripts chunked automatically
- Voice cloning from a consented sample
- 23 languages
Percept — transcription
- Transcript plus SRT and WebVTT with real timings
- Translate a finished transcript and keep its timings
- Deleting a job removes the row and the stored audio
Noema — language
- Eight fixed tasks, never an open chat box
- You see the price before you press the button
- Charged only on success — we absorb the token variance
Subtitles that still line up
Translating a transcript as one blob destroys its segment boundaries. Percept translates segments in numbered batches and maps them back by position, so a translated subtitle file still lines up with the video. WebVTT is what browsers and YouTube take directly.
Transcription runs behind live calls
Your upload is an audiobook or a podcast, not a call turn. Files are staged to object storage and transcribed one job at a time per worker, behind live traffic — so a big upload can never make someone’s phone caller wait.
Cloning is refused without an attestation.
Cloning a voice you do not have permission to use is how a creator tool becomes a deepfake service. We made it a hard gate in code: without an attestation of consent the clone is refused, and the attestation is stored with the voice.
This is the reason a business or a publisher can adopt this at all.
One balance, spent however you like.
The Studio plan carries 8,000 credits a month. What that buys depends on what you point it at.
| Studio rendering | 10 credits / minute |
| Phone calls | 12 credits / minute |
| Speech API (/synthesize, /transcribe) | 4 credits / minute |
| Voice clone | 1,000 credits, one-time |
Clone a voice once. It narrates your episode and answers your phone line.
Studio and the phone agent are the same voice engine drawing on the same credit balance. No creator voice tool can answer a phone. No voice-agent platform gives you a studio.